The Price is Right: How Algorithms Drive Business Goals in Retailing

Author: GumGum

December 15, 2017

This post was contributed and sponsored by GumGum.

Beyond winning individual sales, Eyal Lanxner, CTO and co-founder of Feedvisor says algorithms can give retailers the tools and insights they need to determine retail strategies and set specific business goals. One way Feedvisor is doing this is by helping Amazon retailers win the Buy Box—the coveted ranking that vastly increases the likelihood a consumer will purchase from them.

Lanxner sat down with GumGum CMO Ben Plomion recently to talk about the benefits of artificial intelligence (AI) for retailers.

Ben Plomion: Rules-based technology has served business for a long time. Why then are so many technologies switching to an algorithmic approach?

Eyal Lanxner: There are many variables that influence your performance in a marketplace. Your shipping policy may be fine for one product, but unacceptably slow for another. High storage fees for oversized products may prompt you to adopt more aggressive pricing in order to ensure profitability. Business rules often require you to create a rule—and sometimes more than one—for each product. But that’s practically impossible, especially if you have a large product portfolio.


The Visionary

A one-stop guide to all things AI and computer vision

On top of that, conditions change constantly as new competitors enter the market and others leave it or change their product line. Even if you could keep abreast of all those changes you’d spend all of your time updating your rules. Machine-learning algorithms that continuously learn and adapt to evolving market conditions are better suited to today’s dynamic market. And, they offer you the opportunity to focus your efforts on what you do best: sourcing new products, improving operational efficiencies in your business and other value-added activities.

Plomion: How can algorithms work to promote a specific retailer strategy?

Lanxner: Let’s say for a given product a retailer wants to maximize its profit. It’s very complex because on the one hand, if you sell your product for a very low price you sell more pieces. But how much profit will you gain from that? On the other hand, if you sell at a higher price, each sale will be more profitable. However, you won’t have very many. The challenge is to understand the best pricing point that will provide maximized profit levels.

One thing AI can do is understand the demand curve of the product, meaning how does the price affect the velocity of the orders. Once you know that, it’s possible to draw a profit curve to identify the best price to sell that product at that time.

But the demand curve is sometimes just the starting point, because when maximizing profit, a retailer may also face several constraints. For example, it costs money to store inventory, and those costs may vary depending on the size or complexity of the product. That’s a constraint the algorithm must account for.

Get a handful of marketers in a room and try to count the number of times you hear the word “algorithm.” Algorithm is today what “paradigm” was in the 1990s (or “disrupt” in 2010!).

Unlike paradigms and disruptions, algorithms are real, and they are driving much of the economy. But what are they exactly? How do they work? And how do they drive machine learning?

To find out we tapped Dr. Jean-Luc Thiffeault, a Professor of Applied Math at the University of Wisconsin, Madison.

Plomion: What exactly is an algorithm?

Dr. Jean-Luc Thiffeault: The word has evolved over the years. Originally, mathematicians and computer scientists used the word to describe a very small and well-defined idea, or how to do a mathematical operation efficiently on a computer. Algorithms were things one could publish in a text book, such as The Art of Scientific Computing, which is now in its third edition.

Plomion: Describe an original algorithm. What did they do?

Thiffeault: An algorithm would describe how to do specific tasks, such as how to sort a list of names alphabetically, which sounds trivial, but isn’t when you have a list of millions of names. Prior to an algorithm, the computer would go down the list, and sort two adjacent names as they were encountered. For example, if Jane Smith appeared right before Rachael Adams, the computer would switch the two. But that’s inefficient, because it needs to go back to the top and repeat the process again and again into all names are sorted.

The intelligent idea behind the algorithm is to sort more than just the adjacent names; to swap names much more globally. When the computer encounters a name that starts with Z, put it on the bottom of the list. Teaching a computer to do that isn’t entirely obvious, which is why an algorithm is needed.

For more news and interviews on the intricacies of AI in marketing, subscribe to The Visionary newsletter.